TestAxiom

From requirements to mathematically optimal test cases — with AI intelligence and ISTQB methodology.

TestAxiom is an AI-Powered Test Design Engine that combines Large Language Model intelligence with deterministic, mathematically-grounded test design techniques.

---

Why TestAxiom?

Most AI test generators just throw requirements at an LLM and say "generate tests." The result: bloated, unexplainable, untraceable test suites. Mathematical tools like PICT are precise but require manual parameter extraction.

TestAxiom bridges the gap. Every test case comes with:

• The technique that generated it (EP, BVA, Decision Table, Pairwise, State Transition)
• The mathematical rationale explaining WHY this specific value was chosen
• Full traceability from requirement → technique → test case

---

Quick Start

pip install testaxiom
pip install testaxiom[ai]   # optional: for AI requirement parsing

Python API

from testaxiom import analyze

result = analyze("age", param_type="int", valid_range=(18, 65))
print(result.summary())

Output:

═══ TestAxiom Analysis: age ═══
Techniques applied: EP, BVA
Test cases generated: 7  (87% reduction from 50 exhaustive)

  [EP]  EP-age-001: input=10   → ✗ invalid
         ↳ Equivalence Partitioning: below valid range (18–65). Any value here behaves identically.
  [EP]  EP-age-002: input=30   → ✓ valid
         ↳ Equivalence Partitioning: representative of valid partition.
  [BVA] BVA-age-001: input=17  → ✗ invalid
         ↳ Boundary Value Analysis: just below minimum (18). Classic off-by-one location.
  ...

CLI

testaxiom --param age --type int --range 18 65
testaxiom --param age --type int --range 18 65 --bva-mode 3-value --json

---

Show, Don't Tell — Real Use Cases

1. Free text → pytest tests in 4 lines

import pytest
from testaxiom.parsers import parse_and_analyze
from testaxiom import exporters

# One sentence of requirements → mathematically grounded test cases
results = parse_and_analyze(
    "User age must be between 18 and 65. "
    "Username must be 3–50 characters."
)

# Inject directly into pytest — no manual work
@pytest.mark.parametrize("input_val, expected", exporters.to_pytest_params(results[0]))
def test_age_validation(input_val, expected, app):
    assert app.validate_age(input_val) == (expected == "valid")

2. Decision Table: "If cart > $100 and user is VIP, apply 20% discount"

from testaxiom.engines.decision_table import DecisionTableEngine, Condition, Action

engine = DecisionTableEngine()

conditions = [
    Condition("cart_over_100", "Cart total exceeds $100"),
    Condition("user_is_vip",   "User has VIP status"),
]
actions = [Action("apply_20pct_discount")]

def discount_resolver(conds):
    return {"apply_20pct_discount": conds["cart_over_100"] and conds["user_is_vip"]}

rules = engine.generate_full_table(conditions, actions, action_resolver=discount_resolver)
test_cases = engine.generate(conditions, actions, rules=rules)

# → 4 test cases, covering every combination:
# cart=T, vip=T → discount ✓
# cart=T, vip=F → no discount ✗
# cart=F, vip=T → no discount ✗
# cart=F, vip=F → no discount ✗

3. State Transition: Login flow

from testaxiom.engines.state_transition import (
    State, Transition, InvalidTransition, StateTransitionEngine
)

engine = StateTransitionEngine()

states = [State("logged_out"), State("logged_in"), State("locked")]
transitions = [
    Transition("logged_out", "valid_credentials", "logged_in"),
    Transition("logged_in",  "logout",             "logged_out"),
    Transition("logged_out", "invalid_credentials","locked",    guard="attempts >= 3"),
    Transition("locked",     "admin_unlock",        "logged_out"),
]
invalid = [
    InvalidTransition("locked", "valid_credentials", "Account is locked — must be unlocked first"),
]

test_cases = engine.generate(states, transitions, "logged_out", invalid_transitions=invalid)
# → 4 valid transition tests + 1 negative test, all fully traced

4. Pairwise: Cross-browser/OS matrix (88% reduction)

from testaxiom.engines.pairwise import PairwiseEngine, Parameter

engine = PairwiseEngine()
params = [
    Parameter("browser", ["Chrome", "Firefox", "Safari", "Edge"]),
    Parameter("os",      ["Windows", "macOS", "Linux"]),
    Parameter("lang",    ["en", "he", "ar", "fr"]),
]

cases = engine.generate(params)
# Exhaustive: 4 × 3 × 4 = 48 combinations
# Pairwise:   ~12 test cases — covers every browser×os, browser×lang, os×lang pair
print(cases[0].rationale)
# → "Pairwise test 1/12: covers unique parameter pairs ... 75% reduction"

---

pytest Integration

TestAxiom output plugs directly into @pytest.mark.parametrize:

from testaxiom import analyze, exporters
import pytest

result = analyze("price", param_type="float", valid_range=(0.01, 9999.99))

@pytest.mark.parametrize("price, expected", exporters.to_pytest_params(result))
def test_price_validation(price, expected, pricing_service):
    result = pricing_service.validate(price)
    assert result.status == expected
    # Each test is individually named: BVA-price-001, EP-price-002, ...

to_pytest_params() returns pytest.param objects with auto-generated IDs (EP-age-001, BVA-age-003, etc.) so failures are immediately traceable to their ISTQB technique.

---

AI Requirement Parser

Setup

pip install testaxiom[ai]
export ANTHROPIC_API_KEY="sk-ant-..."   # or set in your environment

Usage

from testaxiom.parsers import parse_requirements, parse_and_analyze

# Extract parameters from free text
specs = parse_requirements(
    "Cart total must be between $0 and $10,000. "
    "Discount percentage: 0–100, must be divisible by 5. "
    "User tier: bronze, silver, gold, platinum."
)

# Or: parse + run full analysis in one call
results = parse_and_analyze("Age must be 18–65", bva_mode="3-value")
for r in results:
    print(r.summary())

Privacy & Security

• Your requirements text is sent to the Anthropic API (Claude) for parsing.
• TestAxiom uses your own ANTHROPIC_API_KEY — no data is routed through TestAxiom servers.
• The system prompt is prompt-cached (Anthropic's caching feature) to reduce latency and cost on repeated calls — no persistent storage of your data.
• If you're working with sensitive specifications, review Anthropic's privacy policy before use. All other TestAxiom engines (EP, BVA, Decision Table, Pairwise, State Transition) are fully local and send no data anywhere.

---

Supported Techniques

Technique	Description
Equivalence Partitioning (EP)	Divides input space into partitions — one test per partition
Boundary Value Analysis (BVA)	Tests at and around partition boundaries (2-value & 3-value modes)
Decision Table	Covers all combinations of business rules (2ⁿ rules for n conditions)
State Transition	Tests valid & invalid transitions in state machines
Pairwise (All-Pairs)	Minimizes combinatorial test sets — covers all value pairs
AI Requirement Parser	Extracts ParameterSpec from free-text requirements via Claude API

All techniques are free and open source (MIT).

---

Architecture

testaxiom/
├── core.py          # Data models (ParameterSpec, TestCase, AnalysisResult)
├── engines/         # Deterministic technique engines — fully local, no AI
│   ├── equivalence_partitioning.py
│   ├── boundary_value.py
│   ├── decision_table.py
│   ├── pairwise.py
│   └── state_transition.py
├── parsers/         # AI layer — optional, requires anthropic package
│   └── ai_requirement_parser.py   # Claude API + tool_use + prompt caching
├── exporters.py     # JSON, Markdown, CSV, Traceability Matrix, pytest params
└── cli.py           # Command-line interface

Key design principles

• Traceability first — every test case includes its technique, partition, and mathematical rationale
• Deterministic — no randomness except Pairwise's greedy search (seeded, reproducible)
• Zero required dependencies — core engines work with no external packages
• AI is optional — pip install testaxiom[ai] only if you want NLP parsing

---

License

MIT

Author

Yaniv (Yaniv2809) — AI-Powered QA Engineer